Screening of potential double perovskite materials for photovoltaic applications using agglomerative hierarchical clustering

TL;DR

This paper introduces a hierarchical clustering method to identify promising double perovskite materials for solar cells, especially effective when data is limited, addressing a key challenge in materials discovery.

Contribution

It presents a novel application of hierarchical clustering for materials screening, improving predictions with small datasets compared to traditional machine learning methods.

Findings

Effective screening of double perovskites for photovoltaics.

Hierarchical clustering outperforms other algorithms on small datasets.

Potential candidates identified for solar cell applications.

Abstract

Data-driven approaches to solve problems in materials science have gained immense popularity in recent times due to their ability to predict unknown material properties and uncover relationships between structure and property. Machine learning algorithms like GBRT, random forest and neural networks have had tremendous success in predicting target properties of materials and design of structures for various applications. However, a major drawback for achieving results within the required accuracy using these algorithms has been the need for large datasets which can be challenging for problems when data is not sufficiently available for training the models. In this work, we propose the use of a hierarchical clustering algorithm which can work considerably better on materials science problems with small dataset constraints. We apply the algorithm to screen out promising double perovskite materials as candidates for solar cells.